Perplexity Turns to Alibaba’s Qwen for Local AI Agent
Perplexity’s new local agent, Portable Computer, runs on Alibaba’s open-source Qwen3.8-27B, a sign of widening adoption by Western AI companies.
Perplexity optimized its PPLX 27B algorithm family on Nvidia DGX Spark hardware so the Qwen model can work locally on file processing, data analysis, and programming tasks. Perplexity, known for AI search and often described as a potential Google rival, has been expanding its agent business from the cloud to local devices; its large-model lineup has also moved from Claude and GPT to Alibaba’s Qwen. Earlier reports said Perplexity had developed an embedding model based on the open-source Qwen3.
Since Qwen3.8-27B was released, it has been widely called a “local Opus 4.6” because of its intelligence density, and it quickly became the preferred model for Perplexity’s local agent business, according to Siliconangle.
Over the past month, Alibaba has open-sourced Qwen3.8-Max, Qwen3.8-27B and Qwen3.8-Flash, achieving state-of-the-art performance and a Pareto-optimal balance of performance and cost. Hugging Face’s summer report described Alibaba’s Qwen series as the base model for global open-source AI. Public reports also show Airbnb relying heavily on Qwen, Pinterest using Qwen for its AI shopping assistant and chatbot, and Reuters building a proprietary model on Qwen.